It's worth noting that I did some googling and couldn't find anything about the architecture, models, LLM, AI stack. If the goal is transparency I'd have expected this to be publicly available to review.

Welcome new friends. 👋

Don't interpret the study as Adopt AI now, there are drawbacks. It should be a guide to think about where it fits in your org

• Developer trust is the barrier to greater usage - Establish an AI usage policy, give people opportunity to experiment but don't force them to use it.
• There are mixed signals coming from other sources/survey - METR stands out and raises questions about the credibility of self reported data

TL;DR
• Mostly self reported data.
• 90% of devs use AI at work.
• Simply adopting tools fix your problems
• AI adoption is a force multiplier - if your delivery practice is good, you'll likely see some benefit. If it's bad, it'll likely make everything worse.

Alternatively get a Mac Mini or studio. Both expensive but fine options.

Overall it seems interesting and useful and something open source tools will likely replicate quickly. There's already an issue in opencode github.com/sst/opencode...

Skills are generally a markdown file called SKILL.md. There are premade examples in a repo here github.com/anthropics/s...

MCP focuses on trying to provide tools to the LLM that it can optionally use. These are more like smaller manuals the LM can reference to improve behavior on certain tasks.

You can have a skill on parsing PDFs as an example - Use pdf kit, ignore table of content, include page numbers - for example